Article
Hard acts to follow: Predecessor effects
on party leader survival
Party Politics
2015, Vol. 21(3) 357–366
ª The Author(s) 2013
Reprints and permission:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1354068812472577
ppq.sagepub.com
Yusaku Horiuchi
Dartmouth College, Hanover, USA
Matthew Laing
Australian National University, Canberra, Australia
Paul ‘t Hart
Utrecht University, Utrecht, The Netherlands
Abstract
In this article, using our original data on party leadership succession in 23 parliamentary democracies, we investigate the
determinants of a party leader’s survival rate: how long he/she remains in office. Unlike previous studies, which focus on
institutional settings of leadership selection or on situational (political, economic and international) conditions at the time of
succession, we propose a perceptual theory of leadership survival, focusing on the expectations of party constituents (or
indirectly, the voting public) who have the power to remove a leader. Specifically, we argue that they ‘benchmark’ their
expectation of a current party leader’s performance by comparing it against their memory of that leader’s immediate
predecessor. Empirically, we show that party leaders who succeeded a (very) long-serving party leader and/or a leader
who had also been the head of government experience lower longevity than others, making these types of predecessor
‘hard acts to follow’.
Keywords
Parliamentary democracies, party leaders, leadership succession, prime ministers, survival analysis
1. Introduction
The leadership position in a political party is a ‘hot seat’
that confers significant power and authority to its occupant. Leaders are key shapers of their party’s electoral
platforms and, if they are also the heads of government,
of public policies. They wield influence in appointments
to party, cabinet, public service, judicial, regulatory and
advisory positions. Furthermore, in recent years, the socalled ‘personalization of politics’ has placed a greater
premium on the personality, style and skills of leaders in
determining the electoral fate of parties and governments
(Aarts et al., 2010; Blondel and Thibault, 2009). Party
leaders who are also the head of government, in particular,
are claimed to have gained power vis-à-vis their party
organizations, cabinet colleagues and line agencies. At the
national level of politics and government in parliamentary
democracies, this trend has recently been referred to,
albeit misleadingly, as the ‘presidentialization’ of prime
ministers (Heffernan, 2003; Poguntke and Webb, 2005;
Samuels, 2002).
Yet, the leadership of a political party, and possibly in its
wake the head of government, is also a high-risk position.
Political leaders operate constantly in the public eye, with all
the responsibilities their position entails, and there are
always plenty of people ready to criticize their performance.
Naturally, there is no shortage of potential competitors,
brooding on how and when to try and take over the top job.
Given the high stakes, fundamentally competitive and
conditional nature of party leadership, political scientists
Paper submitted 01 March 2012; accepted for publication 27 October
2012
Corresponding author:
Matthew Laing, Australian National University, Haydon-Allen Bldg,
Australian National University, Canberra, 0200, Australia.
Email: [email protected]
358
wonder why some party leaders manage to survive in the job
for considerable periods of time, whereas others face relatively quick and often involuntary terminations. This question
is relevant also because a variation in the length of leaders’
survival in office may have important implications. On the
one hand, it may be taken to be a key indicator of the level
of authority and the level of support that leaders (whether in
government or not) enjoy among their principal party constituents – those who have the power to remove them. On the
other hand, very long-serving leaders may stifle innovation
within parties and governments and generate internal conflicts and stalemates without any particular faction having
quite enough power to replace them. It is not hard to imagine
that in these two extreme cases, their eventual successors face
quite different political biotope to secure their own survival.
There is now a considerable corpus of research on political leadership succession. The great majority of studies that
deal with the succession of party leaders in democratic systems examine the effects of institutional rules of leadership
selection and removal on leadership survival. Many such
studies also assess the pros and cons on different leader
selection and removal mechanisms from a number of different perspectives, and examine the impacts of recent institutional changes that have taken place within many parties
(Colomé and Lopez Nieto, 1993; Courtney, 1995; Davis,
1998; Kenig, 2009; LeDuc, 2001; Marsh, 1993; Quinn,
2004, 2005; Stark, 1996; Strøm, 1993; Weller, 1983,
1994). Other empirical studies concentrate on explaining the
dynamics of particular types of successions, such as voluntary or ‘managed’ handovers from an incumbent to an heir
apparent (Bynander and ‘t Hart, 2006, 2007, 2008).
More pertinent to our article are some quantitative studies
examining the determinants of a party leader’s survival rate
– how long he/she remains in office. Seminal studies in this
vein include Bienen and van de Walle (1991), Bueno de
Mesquita et al. (1999, 2003) and Bueno de Mesquita and
Siverson (1995). According to Bueno de Mesquita et al.
(1999), the adaptability of leaders in the face of institutional
boundaries and shifting political coalitions decides their
political longevity. In a subsequent work, Bueno de Mesquita et al. (2003) focus on the role of war and violent conflict in the survival of leaders. Burke (2012) shows that
higher economic growth rates increase the likelihood of
leader survival. By contrast, Chiozza and Goemans (2004)
find that economic performance has a small effect, while
political institutions have a very large effect. They also show
that wars do not necessarily reduce the tenure of leaders, as
compared to crises. By modelling party leadership survival
as a function of electoral performance, Andrews and Jackman (2008) show the vulnerability of leaders to performance
indicators. They also find that those elevated to the leadership close to an election are at a distinct disadvantage.
In this article, we attempt to gain more insight into the
factors that might explain the survival prospects of party
leaders by taking a different approach to those dominating
Party Politics 21(3)
the existing literature. First and foremost, instead of
focusing on institutional settings of leadership selection or
situational (political, economic and international) conditions
at the time of succession, we propose a perceptual theory of
leadership survival, focusing on the expectations of party
constituents who have the power to remove a leader.
Specifically, we argue that they ‘benchmark’ their expectation of a current party leader’s performance by comparing
it against their memory of that leader’s immediate predecessor. We surmise that the constituents remember and recall
things like their predecessors’ political stances and their reputation for efficacy. As a result, some types of predecessors
come out of these comparison processes as much ‘harder
acts to follow’ than others. This study examines the extent
to which this type of paired comparison explains the
differential survival rates of party leaders, regardless of the
prevailing party rules for leader (de)selection.
To do so, we use our new dataset, which includes
information about not only prime ministers (or other ‘effective primary rulers’) but also opposition and minor party
leaders in stable parliamentary democracies as defined by
Lijphart (1999). As we explain later, this allows us to
examine whether survival rates are different depending
on whether leaders succeeded when their party was in
charge of the government or not. To the best of our knowledge, this question has never been examined previously.
Below, we first describe our theory of leadership survival as being contingent upon stakeholders’ perceptions
in greater detail, and articulate a set of hypotheses. We then
explain data and methods used to test our hypotheses in
Section 3. In Section 4, we present and interpret the findings. Specifically, we show that party leaders who succeeded a (very) long-serving party leader and/or a leader
who had also been the head of government experience
lower longevity than others. Section 5 offers key conclusions and suggestions for further research.
2. Hard acts to follow
We propose a perceptual theory of leadership survival. Specifically, we focus on the ways in which the stakeholders
form their expectations for a leader in comparison with
his/her predecessor, and formulate our hypotheses.
2.1. Succeeding a long-serving leader
Whom a leader succeeds does, indeed, matter. This is most
clearly illustrated in the cases of succession of so-called
‘great leaders’ – very long-serving founders or successful
reformers of their parties. Their towering presence in the
top job challenges their followers, critics and other stakeholders to imagine a future without them.
This is typical in non-democratic polities, where there
are few institutional restraints on leadership removal. As a
result, the ageing, illness and/or death of nation-builders and
Horiuchi et al.
dictators often create widespread uncertainty regarding their
succession and regarding the future stability of their regime.
The classic cases of communist party leaders who were also
the heads of government, such as Josip Broz Tito (Yugoslavia, 1943–1980), Mao Zedong (People’s Republic of China,
1949–1976), Kim Il Sung (North Korea, 1948–1994) and
Fidel Castro (Cuba, 1959–2011), illustrate the point vividly.
Also in established democracies, a particular leader’s
longevity in office may give many people the feeling that
things will never be the same without him/her. In the
United States, an entire generation grew up listening to
Franklin D. Roosevelt’s (President, 1933–1945) ‘fireside
chats’, a series of engaging and often soothing addresses
in the dire circumstances of the Great Depression and later
the war. Likewise, during the 1950s and early 1960s in
Australia and Sweden, an entire generation reached adulthood living under one Prime Minister – Robert Menzies
(1949–1966) and Tage Erlander (1945–1968), respectively.
In Germany, the long reign of Helmut Kohl (Chancellor,
1982–1998) earned him the nickname ‘Der ewige Kanzler’
(the eternal chancellor).
Successors to such long-serving leaders face difficulties
that arise out of the sheer fact of their predecessors’
longevity. For one, long-term leaders generally have firm
support bases within the party that have crystallized over
time, leaving successors at a disadvantage in both challenging and reorganizing the long-entrenched party dynamics.
Long-term party leaders also develop a ‘taken for grantedness’ among the party faithful; namely, it is their leadership
style and political profile that dominate the party’s political
memory. In such political/psychological environments, what
we call an ‘attribution effect’ may occur. A party’s success
during a long-serving leader’s tenure is credited to the leader’s credentials, not to exogenous factors. As a consequence,
potential successors face significant challenges in dislodging
constituency expectations, making the case for change
difficult even in the face of declining fortunes of their party.
Historical examples are abundant. In his last years in
office, Sir Winston Churchill (British Prime Minister,
1940–1945 and again 1951–1955) was a faint shadow of
the vigorous anti-appeasement advocate and the wartime
Prime Minister he once had been, but his successor-inwaiting, Anthony Eden (1955–1957), did not risk a major
challenge against the living legend. From 2005 to 2007, the
Australian Treasurer and deputy Liberal Party leader, Peter
Costello, time and again retreated from the brink of openly
challenging the four-term Prime Minister John Howard
(1996–2007) as the party leader, even when Howard’s
public support was evaporating in the run-up to the 2007
election. Likewise, although Helmut Kohl had designated
Wolfgang Schäuble as the next leader of Christian Democratic Union many years in advance, Schäuble never even
considered challenging Kohl to implement the succession,
even in the face of voters’ growing exasperation with Kohl’s
continued and increasingly autocratic reign (Langguth,
359
2009). Schäuble ultimately became the party leader in
1998, after Kohl’s election defeat.
The successors to ‘great leaders’ have to battle with
established party expectations, and face an extraordinarily
tough challenge to be seen and judged in their own right,
rather than in the shadow of their formidable predecessor.
Taking office without the ‘idiosyncrasy credit’ (Hollander,
1978) that comes with a record of survival and achievement
as a party leader, the successors of ‘great leaders’ may have
a much less imposing armour to fend off critics and rivals in
politically lean times for the party.
For example, John Howard’s eventual successor Brendan Nelson lasted less than a year, to be toppled by Malcolm Turnbull who did not survive for much longer
himself. In both cases, the general public and party constituents did not see the kind of sovereign and electorally cunning leadership they had become accustomed to. Likewise,
when Schäuble did become the party leader, he, like Nelson
and Turnbull, inherited a party that was long accustomed to
government and itself being ruled decisively by the ‘big
man’. Though being a long-time Kohl associate and clever
back-room operator, Schäuble lacked Kohl’s huge network
inside the party and simply did not inspire the same level of
internal unity and confidence, and could not survive the
party financing scandal that broke out in 1999 and claimed
his scalp in early 2000.
This is not to say that such successions are inevitably
doomed. Counterexamples are also not hard to find, including
John Major’s (1990–1997) replacement of Margaret Thatcher
(1979–1990) in the British Conservative Party, Olaf Palme’s
(1969–1986) replacement of 23-year Prime Minister Tage
Erlander (1946–1969) in the Swedish Social-Democratic
Party, and Wim Kok’s (1986–2002) replacement of Joop den
Uyl (1969–1986) within the Dutch Social-Democratic Party.
All of these successors went on to win or prolong government
for their parties in their own right.
Generally, however, it seems to suggest that successors
to long-serving leaders face a greater weight of expectations shaped by constituent memories of the ‘golden years’
the party experienced under these ‘great leaders’. We
presume this makes it hard for these successors to assert
authority in their own right, above and beyond the constant
comparisons being drawn between them and their (soon
legendary) predecessors. For these reasons, we propose the
following hypothesis:
Hypothesis 1: Successors to a long-serving party leader
experience lower longevity than other party leaders.
2.2. Succeeding party leaders who are or were
prime minister
The prospects of new leaders may be also influenced by
roles they inherit from their predecessors. The key question
360
here is whether, in becoming a party leader, they also
become the head of government; more specifically, the
prime minister, in our study that focuses on parliamentary
democracies.
One train of thought is that to take over from an incumbent prime minister helps a new leader gain immediate
gravitas by the role as the head of government. This idea
may be similar to the idea of ‘incumbency advantage’,
which is well documented in the literature of electoral
politics (Ansolabehere et al., 2000; Gaines, 1998; Gelman
and King, 1990; Levitt and Wolfram, 1997). Such studies
generally posit an advantage from a combination of more
media exposure and better grounding in the minds of a
leader’s constituencies, not in the least the voters.
This is frequently turned on its head, however. There are
a number of reasons to expect the longevity of successors to
a prime minister to be lower than that of successors to an
opposition leader.
First, the analogy to the incumbency advantage may be
misleading. Consider the circumstances in which parties
choose a successor for an incumbent prime minister. The
first obvious trigger is when a prime minister falls ill or
dies. The involuntary dropping away of the incumbent sets
up a sense of bereavement within the party, which may easily transform into a mythical and phantom presence of the
predecessor among the minds of party members and voters.
Any ‘accidental successor’ (Abbott, 2005), when compared
with his/her predecessor’s canonized past, is doomed to
disappoint stakeholders.
The second, and much more common, trigger is when the
party and the government have got themselves in a really serious mess: internal fighting, entrenched leadership rivalry,
toxic scandals and plummeting polls. Because of the visibility
and importance of the prime minister, and of the importance
of the perception of stability and competence for the party in
government, it is a necessarily high-risk and high-stake decision for a party to replace the head of government mid-stream.
In such circumstances, even hitherto successful government
leaders, such as Thatcher and Jean Chretien (Prime Minister
of Canada, 1993–2003), may be forced out and their successors face a terrible burden of expectations: to learn how to
govern as the new prime minister while at the same time
restoring unity and hope for electoral survival to their divisive, even traumatized, party constituencies.
The difficulties these successors to incumbent prime
ministers face are perhaps irrelevant to whether the transition was voluntary or not. Consider Julia Gillard’s (2000
current) removal of Australian Prime Minister Kevin Rudd
(2006–2010; Prime Minister, 2007–2010). This involuntary
resignation by Rudd is said to have tarnished the party’s
credibility (Coorey, 2010). Thus, from the early stage,
Gillard had to defend herself against strong critics among
party members and voters. Voluntary resignations also tend
to be under a great deal of pressure or in the context of serious
government dysfunction: Harold Macmillan’s (1957–1963)
Party Politics 21(3)
departure as the Prime Minister of the United Kingdom, and
Charles Haughey’s (1987–1992) as Taoiseach of Ireland are
demonstrative of governments in serious trouble rather than
the natural passing of the baton. In both cases, their successors
had to manage party and public expectations in orchestrating
an in-government transition, and such difficult tasks seem
prima facie to have much greater scope for failure than doing
so out of the spotlight of governing.
Considering these, we present the following hypothesis:
Hypothesis 2: Successors to a party leader who also is or
has been the head of government experience lower longevity than successors to an opposition party leader.
The hypothesis pertains, in fact, to two (sub)types of
successions: those of a party leader who is an incumbent
head of government (or hereinafter, IHOG) at the time of
leadership transition, and those of a party leader who is a
previous head of government (PHOG) who has since then
continued or returned as a party leader. We argue that just
like the successors to IHOG, the successors to PHOG are
likely to face challenges to meet the high expectations
among party constituencies. This is because many PHOGs
resign soon after their government’s electoral defeat. While
the election loss still lingers in everyone’s memory, it is a
difficult task for a new leader to build confidence among
party members and to gain support from voters. Alternatively, previous heads of government may still exert influence behind the scenes – in fact many leaders who step
down find it difficult to accept the new directions their successors take. Dom Mintoff (1949–1984), the hero of the
Maltese Labour Party, cast a long shadow over his successors and his presence as a not-so-quiet backbencher would
prove destabilizing for his successors many times over. By
contrast, for a new leader of a party accustomed to losing, it
is relatively easier to meet the lower expectations the party
may have in either the polls or at the ballot box. Jeremy
Thorpe’s (1967–1976) leadership of the UK Liberal Party
might be one such example – a string of by-election wins
and a strong result in 1974 gave him an aura as a saviour
of the party, despite the long-running controversies
concerning his personal life and a lack of gains in parliamentary influence. For an opposition party, small gains
might fortify a leader regardless of other factors.
In short, we argue that Hypothesis 2 is valid regardless
of whether a succession was to IHOG or to PHOG. Both
IHOG successors and PHOG successors have a lower longevity rate than successors to an opposition party leader
without taking the top seat in his/her career.
3. Methods
To test these hypotheses, we use our own dataset covering
23 parliamentary democracies around the world (Australia,
Austria, Barbados, Botswana, Canada, West Germany/
Horiuchi et al.
Germany, Denmark, Great Britain, Greece, India, Ireland,
Israel, Jamaica, Japan, Luxemburg, Malta, The Netherlands,
New Zealand, Norway, Portugal, Spain, Sweden and
Trinidad and Tobago) for the post-World War II period.1
We focus on these countries based on Lijphart’s selection
of 36 democracies (1999). Thirteen countries were not
included in the dataset, however, for several reasons. They
include countries in which a singular party leader is not easily established (France, Italy, Switzerland, Costa Rica and
USA),2 countries that had ceased to be considered ‘free’
according to Freedom House and thus no longer satisfied
Lijphart’s original categorization (Colombia and Venezuela), and countries with which our coders had insufficient
access to English sources (Iceland, Belgium, Luxembourg,
Papua New Guinea, Mauritius and Finland).
The dataset includes 448 leaders from 66 governing and
opposition parties,3 and each observation contains information about when each leader entered and left power, as well
as other relevant personal characteristics.4 Acting or temporary leaders were not counted. There are a small number
of individuals who became a party leader more than once in
non-consecutive periods, but we treat them as independent
observations.5
3.1. Main variables
The dependent variable of our analysis (Incumbent Longevity) is the number of years (more precisely, the number of
days divided by 365) each leader was in power. This duration variable is censored as of 1 October 2009; namely, the
values of our dependent variable for all ‘current’ leaders are
measured up to the censored date.6 It ranges from as little as
47 days in the case of Sir Donald Sangster’s leadership of
the Jamaican Labour Party to an epic 12,852 days (>34
years) in the case of Dom Mintoff’s leadership of the Maltese Labour Party.
Although the arithmetic mean would postulate that leaders on average lead their parties for 5.9 years, the skewed
nature of the longevity distribution means that the large
majority of leaders fall far short of that time period. Specifically, nearly a quarter of leaders (N¼110) did not make it
to the end of a second year in office, and about 40 percent
(N¼177) did not reach beyond a third year. Bueno De Mesquita and Siverson (1995) likewise have shown that the
most perilous period for new leaders is the first two years,
which are hard to survive. At the other end, and equally
remarkable, more than 20 percent of party leaders survived
for ten or more years in office (N¼95). What we see, then,
seems to be a crash or crash-through pattern of party
leadership; namely, most party leaders leave office quickly
but there is a sizeable proportion of leaders with a very long
haul who cannot be ignored.
There are several main independent variables we use to
test our hypotheses. The first (Predecessor’s Longevity) is
the number of years (the number of days divided by 365) a
361
predecessor was in power.7 Our expectation (Hypothesis 1)
is that it has a negative effect on the length of a successor’s
tenure.
The effect, however, may not necessarily be linear;
namely, the effect of an increase in the predecessor’s tenure
from 1 year to 2 years may not be the same as the effect of
an increase from 10 years to 11 years. Considering this
possibility, we also use an alternative specification for the
predecessor’s tenure length. Specifically, we divide the
observations into three groups (almost equally into 149,
149 and 150) and use two dummy variables for the second
and third groups – Medium-Term Predecessor and LongTerm Predecessor. The base category is the first group
(Short-Term Predecessor). The average longevity of predecessors is 1.6 years for the first group, 5.4 years for the second group and 14.7 years for the third group. We expect
negative effects for these dummy variables; namely, compared to leaders who followed a predecessor with a short
tenure, those who succeeded one with medium-term to
long-term tenure are more likely to leave office at any
given time. This negative effect is expected to be particularly large if a predecessor was a long-serving leader.
The other variables for our Hypothesis 2 are concerned
with the type of succession that occurs. We use the
following two dummy variables. The first variable (Previous Head of Government Succession) is coded as 1 if a predecessor had been the head of government (i.e. the prime
minister in our study) at some point previously but was the
leader of a non-governing party at the time of succession,
and zero otherwise. The second variable (Incumbent Head
of Government Succession) is coded as 1 if a predecessor
was the head of government at the time of succession, and
zero otherwise. The base category (No Head of Government Succession) is for cases in which a predecessor has
never led a governing party. The number of observations
is 86, 103 and 259, respectively. As we discussed earlier,
our expectation is that both IHOGs and PHOGs are harder
acts to follow than party leaders who have not held the head
of government position at any point in their career.
3.2. Should we control for institutional changes?
A range of other factors can influence party leaders’ longevity. To control for other determinants, we run a set of
stratified proportional hazards models where ‘strata’ (i.e.
groups) are political parties,8 and estimate the effects of our
independent variables with the strata-specific baseline
hazard.9 Like commonly used OLS regressions with fixed
effects, this approach can effectively control all observable
and even unobservable party-specific variables, as long as
they are constant within the period of our investigation.
Since each party is obviously specific to each country, it
can also control all country-specific variables. Importantly,
these country-specific and party-specific covariates include
a range of institutional settings, which are deemed relevant
362
in shaping the duration of political leaders, such as whether
or not a country has a Westminster system, a two-party
system and/or a fixed election cycle.
Some institutions relevant to the duration of political
leaders, however, may not be constant within the period
of investigation. A potentially important one is the rule for
leadership selection. Generally speaking, the process by
which party leadership is determined has been gradually
democratized and broadened, particularly over the past two
decades (e.g. in the UK, Portugal and Greece). This trend
includes institutional changes to give more opportunities
for removing leaders, such as the introduction of specific
post-election leadership reviews in Canada. The institutional changes, however, also worked in the other direction
in some countries. For example, the reforms in both Italy
and Japan during the 1990s attempted to stabilize leadership, and shifted power and initiative away from constantly
warring factions. There are also a large number of countries
whose leadership selection rules have also remained
relatively unchanged for the duration of the period (e.g.
in Australia, Spain and the Caribbean countries).
Then, an important question is whether and how we control for such institutional changes. We argue that it is not
necessary – or even wrong – to control for any variable
measuring whether the rule of selecting a leader has changed during a specific period for a specific party. This is
because such a change is highly likely to be a consequence
of a predecessor’s longevity, one of our key independent
variables. Adding such ‘post-treatment’ variables into the
analysis introduces serious bias in our estimates (Rosenbaum, 1984). This ‘post-treatment bias’ has not been
sufficiently acknowledged in the social science literature
until recently (see King, 2010).
Let us explain this important point at length. Consider
that there are three leaders succeeding one another – Leader
Z to Leader X, and then Leader X to Leader Y. We want to
estimate whether the longevity of Leader X affects the
longevity of Leader Y. There are three potential cases of
rule changes we should consider.
The first case is a rule change during the period of Leader
Z’s tenure of office. The new rule is likely to affect the
longevity of both Leader X and Leader Y in the same way.
Therefore, this rule change itself cannot explain why and how
the longevity of Leader X affects the longevity of Leader Y.
The second case is a rule change during the period of
Leader X’s tenure of office. This rule change may be a consequence of Leader X’s longevity. For example, a longserving leader may change his/her party’s rule in order to
stay in office for an even longer period. Alternatively, other
factions or would-be leaders may change their party’s rule
to end Leader X’s tenure. This change is also likely to
affect the longevity of Leader Y, Leader X’s successor.
In this case, importantly, the longevity of Leader Y is still
a consequence of the longevity of Leader X, because the
rule change is a consequence of the longevity of Leader X.
Party Politics 21(3)
Methodologically, it is inappropriate to control for a rule
change during the predecessor’s tenure in regression analysis. In a standard regression framework, the estimated
effect of X on Y, given a control variable Z, measures how
much Y will change if X changes and Z does not change. If
Z is, however, at least in part a consequence of X, when X
changes, Z must change as well. Thus, it does not make
sense to assume that Z remains constant when X changes.
This is the essence of the fundamental problem arising from
controlling for post-treatment variables.
There have been some recent attempts by statisticians
and methodologists to deal with this problem and to estimate causal mechanisms in a new framework called ‘causal
mediation analysis’ (e.g. Imai et al., 2011). In our manuscript, however, instead of trying this new approach, by
dropping post-treatment variables, we estimate the total
effect of the predecessor’s longevity on the successor’s
longevity. This total effect is a function of many things that
may happen during the period of a predecessor’s tenure –
namely, variables now called ‘mediators’. We admit that
our statistical analysis does not unpack this ‘black box’
(Imai et al., 2011). This unpacking exercise, however,
requires further theoretical specification and much
advanced methodological sophistication. We leave such
an inquiry for future research.
The third case is a rule change during the period of
Leader Y’s tenure of office. This rule change may be a
function of the longevity of Leader X, his/her predecessor,
or the longevity of Leader Y himself/herself. It may be not
only a consequence but also a cause of the longevity of
Leader Y. For the same reason discussed above, we should
not control whether or not a party’s rule changed during the
period of Leader Y’s tenure as a control variable.
Similarly, many things may happen during Leader X’s
tenure, as well as during Leader Y’s tenure – elections,
political scandals, changes in public attitudes, intra-party
and inter-party conflicts, etc. Again, as they are likely to
be consequences of Leader X’s longevity, we do not add
them as control variables to estimate the effect of Leader
X’s longevity on Leader Y’s longevity.
3.3. Individual-specific or time-specific controls
Although we emphasize the importance of dropping
variables that are at least in part consequences of our main
independent variables, we can control for variables that are
by no means consequences of the main independent variables. First, we add a successor’s age (Age) variable, as
older leaders are at higher risk of being physically unwell
and unable to stay in office.10 Leaders who were older than
the average age at the time of succession (50.5 years old)
survive, on average, for 4.6 years (N¼206), whereas the
average tenure among below-average, younger leaders is
7.0 years (N¼242).11 The younger a leader is, the longer
his/her longevity.
Horiuchi et al.
The second variable (Year) is calendar year in which a
leadership succession occurred. Our data suggest that political leaders’ terms in office have been decreasing in recent
decades. Specifically, the average duration is 8.1 years
(N¼40) among those who succeeded before 1960, 7.6 years
(N¼119) among those who succeeded in the 1960s and
1970s, 6.1 years (N¼181) among those who succeeded in
the 1980s and 1990s, and 2.9 years among those who succeeded after 2000 (N¼50).12 This final finding is no doubt
influenced by the fact that many contemporary leaders in
the database are still in office, and their longevity is determined by the censored date for the dataset rather than what
their actual time in office will be.
Finally, we add a variable (Female) coded as 1 if a successor is female; otherwise it is zero. Although we do not
have a good theory to predict the direction of its effect,
female leaders have a slightly different survival rate than
male leaders. The average length of tenure among female
leaders is slightly shorter (5.7 years, N¼33) than male leaders (6.0 years, N¼415), but we should also note the small
overall proportion of female party leaders.
4. Results
The results from regression analysis based on the stratified
Cox’s proportional hazards model are presented in
Table 1.13 Note that the estimates shown are hazard ratios.
The estimated hazard ratios for the predecessor’s longevity is 1.021 in Model 1 (without control variables) or
1.024 in Model 2 (with control variables); namely, a oneyear increase in the predecessor’s longevity is expected
to increase the hazard ratio by about 2 percent. The magnitude of the effect is small, and the effect becomes significant at the 5 percent level only after other pre-determined
covariates are controlled (Model 2). This is not necessarily
consistent with our expectation.
This puzzling result, however, may be due to the problem in our model specification. As discussed earlier, the
effect of the predecessor’s longevity may not be linear.
Model 3 (without control variables) and Model 4 (with
control variables), which use two categorical variables to
capture the potential non-linearity, show that as compared
to leaders who succeeded short-term predecessors, those
that succeeded long-term predecessors are about 50–59
percent more likely to be out of office at any given elapsed
time since succession: hazard ratio ¼ 1.505 in Model 3 and
1.579 in Model 4. These estimates are significant at the 5
percent level with and without other covariates. The difference between short-term and medium-term predecessors is
small (hazard ratio ¼ 0.886 in Model 3 and 0.989 in
Model 4) and statistically insignificant.
In sum, we are inclined to support our Hypothesis 1, but
in a conditional manner; specifically, the results suggest
that the tenure length of one’s predecessor has a significantly negative effect only when it is very long. Longevity
363
Table 1. The results of survival analysis.
Model
Predecessor’s longevity
1
2
1.021*
[1.68]
1.024**
[2.06]
Medium-term
predecessor
Long-term predecessor
Previous HoG
succession
Incumbent HoG
succession
4
0.886
0.989
[–0.60] [–0.06]
1.505** 1.579**
[2.09]
[2.47]
2.203*** 2.226*** 2.103*** 2.136***
[3.47]
[3.77]
[3.40]
[3.74]
1.898*** 2.176*** 1.758*** 2.023***
[3.28]
Age
Year
Female
Observations
Wald Chi-square
statistic
Log pseudo likelihood
3
448
21.30
[4.00]
1.046***
[4.96]
1.026***
[6.03]
0.725
[–1.09]
448
79.01
[2.85]
448
28.49
[3.62]
1.044***
[4.76]
1.025***
[5.64]
0.673
[–1.41]
448
89.69
–492.84 –468.75 –489.36 –466.58
The dependent variable is Incumbent Longevity. The estimates shown are
hazard ratios. All models are based on a stratified proportional hazards
model where strata are 66 parties. The robust clustered z statistics are in
parentheses, where clusters are also 66 parties. HoG ¼ Head of
Government. ***p<0.01, **p<0.05, *p<0.10.
may constitute a threshold effect; namely, it only becomes
a noticeably complicating factor once it has exceeded a certain point. It may take a decade or more of a particular party
leader to reign before he/she starts to become a kind of iconic benchmark that complicates the life of a successor. Further examination of how and why predecessor longevity
becomes relevant in the longer term must be left for future
research.
The results for two categorical variables measuring
the types of leadership successions Previous Head of
Government Succession and Incumbent Head of Government Succession show that compared to leaders who
succeeded leaders without any experience as the head
of government, the hazard ratio of those who succeeded
heads of government double at any given time, ranging
from 1.758 to 2.226. The effects are highly significant at
the 1 percent level. These results are consistent with our
Hypothesis 2.
Interestingly, the comparisons of hazard ratios for these
two variables suggest that whether a predecessor was the
head of government at the time of succession or not does
not matter. In each estimated model, the difference between
the two hazard ratios is statistically insignificant at any
conventional level. These results are consistent with our
initial expectation: Both PHOGs and IHOGs are (roughly)
equally difficult to follow for successors.
364
The estimated hazard ratios for control variables also
show some interesting patterns. First, younger party leaders
are more likely to stay in office for a longer period. We can
only speculate as to why this is: voters in parliamentary
democracies may be more likely to show tolerance to
younger leaders in a way that they are not with wellworn members of the establishment. Alternatively, these
young leaders may be more skilful in interpreting voters’
voices and reflecting them in policy process. The answer
may be situational too; namely, young leaders tend to be
reformers who signal generational change, which often
means necessary renewal for a tired party.
Another finding is that the average leadership tenure has
been on the decline in recent decades, which may be related
to increasing pressures to being seen to ‘score’ and yet to
avoid scandal generated by the proliferation and intensity
of the electronic media’s coverage of politics. The growing
density of polling and the ‘perpetual campaign’ are particularly problematic for opposition party leaders, who find
themselves with shorter and shorter periods in which to
establish their credentials before being replaced.
Furthermore, there is no statistically significant difference between the tenure length of female leaders and that
of male leaders. Female politicians likely have different
supporting bases than male politicians and face a wide
array of obstacles to gaining the party leadership (as evidenced by the very small number of female party leaders,
even in contemporary decades). Yet, once a female politician becomes a party leader, it seems that there is no significant difference in their longevity, though this finding is
perhaps skewed by a couple of highly successful female
leaders (e.g. Thatcher and Merkel).
5. Conclusions
Leadership successions are seldom completely smooth
affairs, no matter how well they are planned for or how
thoroughly successors have been groomed. However,
successors to a very long-term leader and successors to the
incumbent or former head of government face particular
challenges in securing a mandate of their own. Our results
suggest that these challenges are not trivial ones and constitute a ‘hard acts to follow’ effect, which complicates and
intensifies risk for certain leadership succession scenarios.
These results also suggest further room to refine and test
our perceptual theory of leadership transition. Previous
studies of party leadership succession have mainly focused
on the roles of institutional settings and situational conditions. We do not deny them as important. Yet, we argue that
one should also consider the psychology of stakeholders
who have power to remove a party leader. They form their
expectations of the performance of a new leader in light of
their experiences with his/her immediate predecessor. As a
consequence, particularly powerful or otherwise iconic predecessors become hard acts to follow. We believe that an
Party Politics 21(3)
important research agenda in the literature of political leadership is to combine institutional and situational factors
intertwined with psychological factors to explain party
leader longevity and succession.
Poorly timed and poorly executed leadership changes
are both common and deeply problematic for political
parties. In the worst cases, they can lead to complete destabilization within a party, incessant leadership speculation
and rivalry, and protracted abandonment by the voters.
More research is needed as to why the ‘hard acts to follow’
effect exists, and how leaders and parties can minimize the
odds of succession traumas.
Funding
This research received no specific grant from any funding agency
in the public, commercial or not-for-profit sectors.
Notes
1. Party leaders who were in office as of 1945 are included in
our dataset. Thus, in some cases those leaders might have
assumed office as early as 1926.
2. This was usually because the system of government divides
effective political leadership (e.g. USA) or because officeholding and effective leadership may not coincide (e.g. Italy
pre-mani pulite).
3. We include parties that had formed government (either outright or in coalition) during the period studied. We exclude
some parties meeting this criterion due to limited
information.
4. We collected information from a range of sources, including
party and government websites, Keesing’s World News
Archives, the New York Times and other international news
outlets, national political histories, and some of the existing
case and comparative studies of leadership succession (e.g.
Calvert (1987), Davis (1998) and various articles in a special
issue on party leadership; for example, see European Consortium for Political Research (1993)).
5. As a robustness test, we estimated our models by dropping
these observations, but our conclusion drawn from these
results is unchanged.
6. In our dataset, there are 63 leaders still in power as of 1
October 2009.
7. It is adjusted to the difference from the minimum.
8. We use the Cox proportional hazard model, as there is no
prior belief about the functional form of the baseline hazard
(Box-Steffensmeier and Jones, 2004: 66; Andrews and Jackman, 2008: 667).
9. As we assume that observations are independent across political parties but not necessarily within each party, we also estimate clustered robust standard errors of coefficient estimates,
where clusters are the strata.
10. It is adjusted to the difference from the minimum.
11. As a robustness check, we tried to use this dichotomized variable (whether a successor’s age is age at the time of succession
was above the average) instead of Age, but the estimated effects
Horiuchi et al.
of our main causal variables (i.e. their direction, the magnitude
and the level of significance) did not change substantially.
12. These simple averages may suggest a non-linear effect on the
dependent variable. Adding squared and cubed terms to
account for a non-linear effect did not yield substantially
different results, either.
13. An important assumption in Cox proportional hazards models
is that the hazard ratio is constant at any given time. A
common way to test this assumption is to run a regression
of residuals on time and test the null hypothesis that the slope
is zero for individual covariates and jointly zero for all covariates (Grambsch and Therneau, 1994). The test results
(available upon request from the authors) show that all control variables, individually and jointly, have no statistically
significant effects, at the 10 percent level of significance,
on the variation in residuals.
References
Aarts K, Blais A and Schmitt H (2010) Political Leaders and
Democratic Elections. Oxford: Oxford University Press.
Abbott P (2005) Accidental presidents: Assassination, resignation
and democratic succession. Presidential Studies Quarterly
35(4): 627–645.
Andrews J and Jackman RW (2008) If winning isn’t everything,
why do they keep score? Consequences of electoral performance for party leaders. British Journal of Political Science
38(4): 657–675.
Ansolabehere S, Snyder J and Stewart C (2000) Old voters, new
voters, and the personal vote: Using redistricting to measure
the incumbency advantage. American Journal of Political Science 44(1): 17–34.
Bienen H and van de Walle N (1991) Of Time and Power: Leadership Duration in the Modern World. Stanford, CA: Stanford
University Press.
Blondel J and Thibault J-L (2009) Political Leadership, Parties
and Citizens: The Personalisation of Leadership. London:
Routledge.
Box-Steffensmeier J and Jones BS (2004) Timing and Political
Change: Event History Modeling in Political Science. Ann
Arbor, MI: University of Michigan.
Bueno de Mesquita B and Siverson RM (1995) War and the survival of political leaders: A comparative study of regime types
and political accountability. American Political Science
Review 89(4).
Bueno de Mesquita B, Morrow JD, Siverson RM and Smith A (1999)
Policy failure and political survival – the contribution of political
institutions. Journal of Conflict Resolution 43(2): 147–161.
Bueno de Mesquita B, Smith A, Siverson RM and Morrow JD (2003)
The Logic of Political Survival. Cambridge, MA: MIT Press.
Burke P (2012) Economic Growth and Political Survival. The B.E.
Journal of Macroeconomics 12(1).
Bynander F and ’t Hart P (2006) When power changes hands: The
political psychology of leadership succession in democracies.
Political Psychology 27(5): 707–730.
365
Bynander F and ’t Hart P (2007) The politics of party leader
survival and succession: Australia in comparative succession.
Australian Journal of Political Science 42(1): 47–72.
Bynander F and ’t Hart P (2008) The art of handing over: (Mis)managing party leadership successions. Government and
Opposition 43(3): 385–404.
Calvert P (1987) The Process of Political Succession. Basingstoke: Macmillan.
Chiozza G and Goemans HE (2004) International conflict and the
tenure of leaders: Is war still ex post inefficient? American
Journal of Political Science 48(3): 604–619.
Colomé G and Lopez Nieto L (1993) The selection of party leaders in
Spain. European Journal of Political Research 24(3): 349–360.
Coorey P (2010) Rudd sucking oxygen from Gillard’s campaign.
Sydney Morning Herald July 22. http://www.theage.com.au/
federal-election/rudd-sucking-oxygen-from-gillards-campaign20100722-10lsm.html
Courtney JC (1995) Do Conventions Matter? Choosing National
Party Leaders in Canada. Montreal: McGill-Queens University
Press.
Davis JW (1998) Leadership Selection in Six Western Democracies. Westport, CT: Greenwood Press.
European Consortium for Political Research (1993) Special Issue.
European Journal of Political Research, 24(4).
Gaines B (1998) The impersonal vote? Constituency service and
incumbency advantage in British elections, 1950–1992. Legislative Studies Quarterly 23(2): 167–195.
Gelman A and King G (1990) Estimating incumbency advantage
without bias. American Journal of Political Science 34(4):
1142–1164.
Grambsch P and Therneau TM (1994) Proportional hazards tests
and diagnostics based on weighted residuals. Biometrika
81(3): 515–526.
Heffernan R (2003) Prime ministerial predominance? Core executive politics in the United Kingdom. British Journal of Politics and International Relations 5(3): 347–372.
Hollander EP (1978) Leadership Dynamics: A Practical Guide to
Effective Relationships. New York: Free Press.
Imai K, Keele L, Tingley D and Yamamoto T (2011) Unpacking
the black box: Learning about causal mechanisms from experimental and observational studies. American Political Science
Review 105(4): 765–789.
Kenig O (2009) Democratization of party leadership selection: Do
wider selectorates produce more competitive contests?
Electoral Studies 28(2): 240–247.
King G (2010) A hard unsolved problem? Post-treatment bias in
big social science questions. Hard Problems in Social Science
Symposium. Harvard University, Cambridge MA.
Langguth G (2009) Kohl, Schroder, Merkel: Machtmenschen.
Munich: Deutscher Taschenbuch Verlag.
LeDuc L (2001) Democratizing party leadership selection. Party
Politics 7(3): 323–341.
Levitt SD and Wolfram C (1997) Decomposing the sources of
incumbency advantage in the U.S. House. Legislative Studies
Quarterly 22(1): 45–60.
366
Lijphart A (1999) Patterns of Democracy. New Haven, CT: Yale
University Press.
Marsh M (1993) Selecting party leaders in the Republic of
Ireland: Taking the lid off party politics. European Journal
of Political Research 24(3): 295–316.
Poguntke T and Webb P (2005) The Presidentialization of Politics: A Comparative Study of Modern Democracies. Oxford:
Oxford University Press.
Quinn T (2004) Electing the leader: The British Labour Party’s
electoral college. British Journal of Politics and International
Relations 6(3): 333–352.
Quinn T (2005) Leashold or freehold? Leader eviction rules in the
British Conservative and Labour parties. Political Studies
53(4): 793–815.
Rosenbaum PR (1984) The consequences of adjustment for a concomitant variable that has been affected by the treatment.
Journal of the Royal Statistical Society, Series A (General),
147(5): 656–666.
Samuels DJ (2002) Presidentialized parties. Comparative
Political Studies 35(4): 461–483.
Stark LP (1996) Choosing a Leader: Party Leadership Contests in Britain from Macmillan to Blair. London:
Macmillan.
Strøm K (1993) Competition ruins the good life: Party leadership
in Norway. European Journal of Political Research 24(3):
317–347.
Weller P (1983) The vulnerability of prime ministers: A comparative perspective. Parliamentary Affairs 36(1): 96–117.
Party Politics 21(3)
Weller P (1994) Party rules and the dismissal of prime ministers:
Comparative perspectives from Britain, Canada and Australia.
Parliamentary Affairs 47(2): 133–143.
Author biographies
Yusaku Horiuchi is Associate Professor of Government and holds
the Mitsui Chair in the Study of Japan at Dartmouth College. He is
the author of Institutions, Incentives and Electoral Participation in
Japan: Cross-Level and Cross-National Perspectives (Routledge
2005), and has published articles in American Political Science
Review, American Journal of Political Science, Journal of Politics,
British Journal of Political Science and Comparative Political
Studies, among others. His recent work includes the applications
of natural experiments in various electoral settings.
Matthew Laing is a postdoctoral researcher from the Australian
National University, specializing in leadership, comparative politics and American government and history. His current work concerns the development of prime ministerial government in
Australia, and has recently published on the subject of Stephen
Skowronek’s political time as it concerns Australian prime ministers and the presidency of Barack Obama.
Paul ‘t Hart is professor of public administration at Utrecht University and associate dean of the Netherlands School of Public
Administration. He is co-editor of Political Psychology. His own
research is on political and administrative leadership, crisis management, public accountability and policy evaluation. His latest
book is Understanding Prime Ministerial Performance (Oxford
University Press, 2013).